Why America Must Dominate Open-Source AI
China’s open-source AI models are spreading globally. To preserve technological leadership, the United States must lead not only in capabilities but in openness.
When Z.ai released its open-weight GLM-5.2 model last month, performing nearly on par with the leading frontier models (GPT, Claude, and Gemini), artificial intelligence (AI) firms everywhere got a wake-up call. America’s top frontier AI companies crush the competition in Large Language Model (LLM) capabilities and revenue, but China’s open models are beginning to seed the world and rapidly regain ground. The key question won’t be who wins the next benchmark, but who supplies the stack the world builds on. Will it be the free world of liberal democracies or the despotic surveillance-state regimes?
America Leads in Closed AI
Considering most foundational research and development took place on US soil, it’s no surprise that American companies have held a lead in AI. Anthropic’s Claude and OpenAI’s GPT lead in capabilities, intelligence, and user adoption, while Google’s Gemini has an edge in native video understanding and generation, as well as in processing audio and video natively. xAI’s Grok models have real-time data that no other platform has, thanks to X and the conversations happening there 24/7. That data translates to knowledge.
But all these models suffer from the same fundamental problem. While they excel at coding and research aimed at boosting productivity, each remains a closed-source model that functions effectively as a black box to users.
No customization, no ability to self-host or own the data, and no ultimate control if users are cut off, as we saw recently with the government-ordered suspensions of Claude’s Mythos 5 and Fable 5, and the restricted rollout of GPT-5.6 over “national security” concerns.
If you’re a healthcare company that uses a Claude model to build dashboards and tools for hospitals, that’s a problem. If you used a proprietary model to build a fraud-detection pipeline or cybersecurity workflow, losing access is a major operational risk.
China Is Winning the Open-Source AI Race
China’s top models, including DeepSeek, GLM, and Qwen, on the other hand, are released with open weights, allowing anyone with sufficient hardware or compute to download and run them and customize them to their heart’s content. China, the model for 21st-century AI-backed authoritarian governance, is besting the liberal West on open-source.
For an entrepreneur or startup, open-source software reduces overreliance on high-cost tokens associated with proprietary models and frees them to scale the technology and find new use cases. To handle sensitive health and banking information, businesses and users can run open models in encrypted environments that only they can see and control.
Can you say the same if you gush all your secrets to a closed-source LLM?
Open-weight proliferation poses real risks and trade-offs, but ceding the global stack to China poses even greater ones, especially for innovators.
As Palantir CEO Alex Karp said last week, business users are growing skeptical of the frontier companies because they want “control over their compute, their models, their data stack, and their alpha. They want to know they own the means of production and it’s not being transferred to someone else.”
That’s very reasonable on the part of businesses.
Why Open-Source AI Matters
American companies have released strong open-source and open-weight AI projects: Meta’s Llama, Google’s Gemma, xAI’s open Grok releases, and OpenAI’s GPT-OSS. But none approach flagship capability.
Open-source AI matters because it defines the ecosystem. It provides the foundation and sets parameters for the next layers of progress in AI, just like Americans did when we built the open infrastructure of the internet and early AI: BSD Unix, PostgreSQL, Firefox, TensorFlow, and PyTorch.
Harvard researchers estimate that if open-source software disappeared tomorrow, firms would have to spend $8.8 trillion to replace what they now use for free, much of it built in America’s own research labs, companies, and developer culture.
When AI innovation is built on American rails, that means American principles are baked in from the start. When compared to a quasi-communist surveillance state with geopolitical ambitions, this matters.
Open AI Leadership Requires Competition, Not Regulatory Capture
Americans should make money, start companies, and protect their investments and IP.
But as we’ve seen recently, frontier AI companies have become political actors of their own making, sparring with the Pentagon, reportedly promising equity to the government in exchange for less regulation, and having their models shut down by federal agencies after making hyperbolic claims of an AI-job-loss apocalypse.
This is a regulatory capture strategy.
The better way is to champion the development and spread of open tech for the world. Remove the regulatory hurdles that favor incumbents, grant innovators the flexibility to launch and adapt models, preempt state rules, and ensure that export controls do not stunt American tech leadership.
America Needs an Open AI Strategy
We can celebrate OpenAI, Anthropic, and xAI for winning. But if America ultimately wants to defeat its geopolitical rivals and be a dominant economic power, it can’t just sell black-box software. We must provide a better technology stack to build on—and that’s not charity—it’s power.
Yaël Ossowski is deputy director at the Consumer Choice Center and a fellow at the Bitcoin Policy Institute. He studied at Concordia University in Montréal and the University of Vienna and received an MA in Philosophy, Politics, Economics (PPE) at the CEVRO Institute in Prague.
Originally published in The National Interest (archive #1, #2).